Pricing and Control of the Internet

Sunday, March 7, 2004 - 3:15pm - 4:30pm
Keller 3-180
Rayadurgam Srikant (University of Illinois at Urbana-Champaign)
In the first part of the tutorial, we will present a general introduction to the architecture of the Internet. Various protocols for scheduling, admission control, routing and congestion control will be introduced. We will then focus our attention on TCP, the widely-used protocol for file transfer in the Internet today. Jacobson's TCP congestion control algorithm has been remarkably successful in regulating file transfers and facilitating the phenomenal growth of the Internet over the last decade. This congestion control mechanism was designed for networks where the required data rate per user is small (less than one Mbps) and the round-trip times are small (of the order of a few milliseconds). However, access speeds, application requirements and file transfer distances continue to increase. Using simple tools from queueing theory and delay-differential equations, we will illustrate the need to redesign the congestion management mechanisms in the Internet to efficiently deliver high data rates over long distances.

In the second part of the tutorial, we will concentrate on pricing and control mechanisms that have recently led to the design of scalable TCP protocols. Starting with Kelly's model of resource allocation in a heterogeneous Internet, it will be shown that congestion management can be viewed as a distributed algorithm for fair resource allocation in the Internet. We will illustrate the use of tools from convex optimization, stochastic processes and control theory in designing congestion control mechanisms at the end users and congestion indication mechanisms at the routers that deliver an efficient loss-free, delay-free service over the Internet.